AI Fruit has Take a Long Time to Ripen

With all the present hype about generative AI and AI in general, it is worth considering that important new technologies that become widely adopted often take quite a long time to reach  “ubiquity.” 

And artificial intelligence is the poster child for expected major technology that has taken many decades to reach “awareness,” and is only in the early stages of “adoption.” Granted, there are very high hopes for applied use cases, such as generative AI. 

Some might point to the Eliza chatbot, created in 1966, as a starting point. Assuming we can say that generative AI is entering the early phases of adoption, then Eliza, a rule-based system that could simulate human conversation, represents a 57-year gestation.

To the extent that generative AI is a form of neural network, and if we date neural network research to the early 1980s, with a working prototype created in 1987, then generative AI has been in development for 36 years. 

To the extent that generative AI is an application of generative adversarial networks in 2014, then generative AI to support use cases such as Image, text, music, video generation; gaming or drug discovery might be said to have been nearly a decade in the making.

And while AI has been an unusually slow developing capability, even lots of other innovations have taken decades to reach ubiquitous use. 

With the caveat that we can argue about when “adoption” first happened, it can easily take a decade before any important and valuable new technology is generally used by four out of five people. “Decades” is not an unusual length of time before that happens. 

Innovation	Year launched	Year reached 10% adoption	Year reached 80% adoption	Years to reach 80% adoption
Personal computers	1975	1983	1995	20 years
The internet	1983	1993	2000	17 years
Cell phones	1983	1994	2005	22 years
Graphical user interfaces (GUIs)	1984	1989	1997	13 years
The World Wide Web (WWW)	1989	1994	2000	11 years
Email	1971	1993	2003	32 years
Search engines	1990	1998	2007	17 years
Online shopping	1994	1999	2008	15 years
Social media	1997	2005	2013	16 years
Smartphones	2007	2010	2017	10 years
Cloud computing	2008	2013	2019	11+ years
Big data	2010	2015	2020	10 years
Artificial intelligence (AI)	1956	2012	2022	66 years+
Virtual reality (VR)	1992	2016	2022	30 years+
Augmented reality (AR)	1990	2016	2022	30 years+

AI is important, most of us agree. But the commercial fruit will take some time to ripen.

Why Consumers have "Satisfaction" Issues with "Intangible" Products

Perhaps all “intangible” products delivered as “services” suffer from a common problem: consumers have a hard time evaluating the quality of the provided services. And that might easily lead to lower satisfaction ratings. 

Intangible product customer acquisition is one matter, but customer retention arguably is quite a bit harder, as the customer experience is everything for an intangible product (education, legal or financial advice, connectivity services, public transportation, temporary lodging).

Perhaps that is why “service” industries have lower customer satisfaction ratings than “product” industries. Though particular scores might vary from study to study, the general pattern is clear enough: product satisfaction tends to be higher than “service” satisfaction. 

Industry	Type	Customer Satisfaction Rating
Product Industries
Automobiles	Product	75%
Electronics	Product	78%
Food & Beverages	Product	79%
Clothing & Footwear	Product	76%
Service Industries
Banking & Financial Services	Service	72%
Healthcare	Service	74%
Retail	Service	73%
Travel & Tourism	Service	71%
Intangible Services
Mobile & Fixed Communication Services	Service	70%
Internet Access	Service	68%
Subscription TV	Service	67%

Intangible products tend to be highly people-intensive in their production and delivery, which means consistency is a challenge. 

The more people-intensive a product, the more room there is for personal discretion, idiosyncrasy, error, delay or other “experience” problems. 

Colloquially, customers have no way of knowing whether the service they are buying represents high value, compared to cost, or not. 

In many cases, the context is essentially binary: one has a cell phone signal or not; internet access is working, or not; electrical power is available or not. At the margin, customer service becomes crucial, especially when services are degraded and a fix is required. 

Consumer service customer attrition can be quite high, as much as 10 percent per month for streaming video subscription services, for example. At rates that high, half of the members of a cohort have abandoned service after about six months.

In decades past, churn rates for many consumer connectivity services was as high as three percent a month, representing nearly 100 percent turnover of a single cohort over three years. 

At the same time, customer satisfaction scores for most connectivity products is quite low. And customer satisfaction scores for virtually all industries fell significantly during the Covid pandemic. 

But connectivity service customer satisfaction scores, as measured by the American Customer Satisfaction Index, have been at the bottom end of scores for all industries, for many decades, despite significant improvements on many metrics of network reliability, availability, billing accuracy and customer service. 

And you might think satisfaction scores would be highest for internet service (since it is, as we tend to agree, an “essential” service), compared to voice services or subscription video. In fact, ISPs tend to score the lowest on ACSI surveys of any industry, just below subscription video services, which historically have ranked at the very bottom of ACSI surveys. 

To be sure, mobility service ranked highest of the connectivity services, but still in the bottom 20 percent of all products and industries. 

So one wonders whether all subscription services will tend to suffer from lowish evaluations, either because customers dislike the feeling of being “locked in” to a supplier; because some subscription services are inherently more complex, in terms of availability; or because customers are simply reminded every billing period that they still are paying for a product. 

Also, some subscriptions are more expensive than one-time purchases of the same products. And many subscriptions also seem to come with predictable annual price increases as well. 

Home Broadband Improvements Now Come at Twice Moore's Law Rates

In one sense, we should not be so worried that internet access providers will be able to keep increasing capacity to match end user demand. Consider the rates of improvement for computing and storage, using Moore’s Law, rates of improvement for fixed network internet access and mobile internet access. 

The computing baseline would be a doubling of capability every two years. Using that standard, communication networks lagged Moore’s Law rates of improvement in the 1980s. Fixed networks reached parity in the decade of the 2000s, while mobile networks reached parity in the 2020s. 

Decade	Moore's Law (doubling time)	Home Broadband Speed (doubling time)	Mobile Internet Speed (doubling time)
1980s	2 years	4 years	10 years
1990s	2 years	3 years	6 years
2000s	2 years	2 years	4 years
2010s	2 years	1.5 years	3 years
2020s	2 years	1 year	2 years

In fact, the rate of fixed network capacity increases now exceeds that of Moore’s Law. In other words, fixed network capacities are  improving twice as fast as computing capabilities. And while mobile networks have generally been more bandwidth challenged than the fixed networks, mobile capacity gains now equal Moore’s Law rates of improvement. 

Over time, that results in exponential rates of change. Home broadband speeds, in fact, do correspond with Edholn's Law or Nielsen's Law\ of bandwidth increase. 

Edholm’s Law states that internet access bandwidth at the top end increases at about the same rate as Moore’s Law suggests computing power will increase. Nielsen's Law essentially is the same as Edholm’s Law, predicting an increase in the headline speed of about 50 percent per year. 

Nielsen's Law, like Edholm’s Law, suggests a headline speed of 10 Gbps will be commercially available by about 2025, so the commercial offering of 2-Gbps and 5-Gbps is right on the path to 10 Gbps. 

source: NCTA  

To be sure, some access providers worry about capital investment costs. But, historically, internet service provider revenues and higher speeds have moved largely in tandem, even if cost per bit metrics show a steady trend towards lower per-unit cost. 

The overall trend for internet access is higher consumption at lower unit costs. In other words, we use more data, consumed at higher speeds, but also at more-affordable costs over time, adjusting for inflation and hedonic improvements. 

It is hard to answer the question “have home broadband prices risen since 2009?” without using hedonic adjustment and also adjusting for inflation. The Bureau of Labor Statistics uses hedonic adjustment to track producer prices for home broadband, for example, since speed and other attributes change over time. 

The rationale is that a dial-up internet connection is not a  comparable service to home broadband at various speeds (10 Mbps, 100 Mbps, 1 Gbps, for example). Since prices tend to stay about the same over time while speeds have increased for the “most bought” tiers of service, BLS adjusts prices to account for quality improvements. 

source: Bureau of Labor Statistics 

The bottom line is that ISPs do not seem to be faced with business scenarios where consumer demand cannot be supplied.